2026 Coding Plan Panorama Survey: Who is using fixed monthly fees to redefine the price of AI programming?
In the spring of 2026, seven major domestic cloud manufacturers will simultaneously enter the AI programming subscription track. From Alibaba Cloud's 7.9 month of the head of state to Cursor/Claude Code's overseas model-how should developers choose?
When Alibaba Cloud raised the price of the Coding Plan for the first month to 7.9 yuan, the price war was no longer a question of who was cheaper-it was why you were paying the bill.
In the spring of 2026, the seven major domestic cloud manufacturers will enter the "AI Programming Subscription Package" track almost simultaneously. From Smart Spectrum taking the lead in testing the water to Alibaba Cloud exploding the price of cabbage, to Tencent Cloud, Volcano Ark, Baidu Qianfan, and Unicom Cloud entering the market in a row-developers suddenly found themselves in front of more than a dozen seemingly similar package menus, with prices ranging from 7.9 to 469 yuan, with supported models ranging from 4 to 20.
Choice anxiety is more intense than Token anxiety.
first figured out one thing: What exactly is a Coding Plan?
Coding Plan is not an API. This is the most important prerequisite.
The traditional method is: register an API Key → charge by Token → invoke the model. You can use it in any scenario-Web applications, backend services, automated scripts, Telegram Bot, whatever.
The Coding Plan is: pay a fixed monthly fee → get an exclusive Key → It can only be used interactively in IDE programming tools . It is essentially an "AI Programming Assistant Membership Card" rather than a universal API quota.
This difference determines: If you want to use the Coding Plan Key to run Dify workflows, build Telegram Bot, or do any background automation tasks-if you violate the rules, you will be banned.
Domestic Coding Plan Panoramic Comparison
As of April 2026, at least 10 domestic manufacturers have launched Coding Plans. Prices are surprisingly convergent (Lite ¥40/month, Pro ¥200/month are the mainstream prices), and the differences are hidden in three dimensions: model quality, quota consumption rules, and tool compatibility.
Detailed Parameter Sheet
| platform | Lite monthly fee | Pro monthly fee | First month discount | core model | Lite Monthly Request | Pro Monthly Request | compatible instruments |
|---|---|---|---|---|---|---|---|
| Alibaba Cloud Hundred Refining | ¥40 | ¥200 | ¥7.9/¥39.9 | Qwen3.5-Plus, GLM-5, Kimi-K2.5, MiniMax-M2.5 | 18,000 | 90,000 | Cursor, Claude Code, Cline, OpenClaw, etc. 20 + |
| Tencent Cloud | ¥40 | ¥200 | ¥7.9/¥39.9 | HY 2.0 Instruct, GLM-5, Kimi-K2.5, MiniMax-M2.5 | 18,000 | 90,000 | Cursor, Claude Code, Cline, CodeBuddy |
| Volcanic Ark | ¥40 | ¥200 | ¥9.9/¥49.9 | Doubao-Seed-2.0-Code, GLM-4.7, DeepSeek-V3.2, Kimi-K2 | 18,000 | 90,000 | Claude Code, Cursor, Cline, Codex CLI |
| Baidu Qianfan | ¥40 | ¥200 | ¥7.9/¥39.9 | GLM-5, MiniMax-M2.5, Kimi-K2.5, DeepSeek-V3.2 | 18,000 | 90,000 | Claude Code, Cursor, Cline |
| Unicom Cloud | ¥40 | ¥200 | Has been exempted for 1 month | GLM-5, MiniMax-M2.5, Qwen3.5, DSV3.X | 18,000 | 90,000 | OpenCode, Claude Code, OpenClaw |
| Intelligent GLM | ¥49 | ¥149/¥469(Max) | no | GLM-5, GLM-5.1 | ~80 times/5h | ~400 times/5h | Cursor, Claude Code, Cline |
| Kimi (Moonshot) | ¥19 | ¥49-99 | - | Kimi-K2.5-Thinking | Press gear | Press gear | Limited tools |
| MiniMax | ¥29 | ¥49-119 | - | MiniMax-M2.5 | No weekly limit | No weekly limit | Multi-tool compatibility |
| Wuwen Core Dome | ¥19.9 | - | - | multi-model | Close to Bailian Lite | - | Multi-tool compatibility |
quota consumption trap : one "question" ≠ one "request." In Cursor/Claude Code, an edit action triggers 5 - 30 model calls. Lite's 18,000 sounds like a lot, but it's actually only enough for 600 - 3,000 human interactions.
Foreign Coding Tools Subscription Comparison
Overseas markets are taking another path-not selling API quotas, but selling IDE products themselves.
| tool | free | Pro | senior | model | billing mode |
|---|---|---|---|---|---|
| Cursor | 50 slow requests | $20/month | $60(Pro+)/$200(Ultra) | Sonnet 4.6, GPT-5.4, Gemini 3 Pro | Credit Pool ($20 = ¥ 20 pool) |
| GitHub Copilot | 50 messages +2000 completions | $10/month | $39(Pro+) | GPT-4o, Claude Sonnet, Opus 4.6(Pro+) | Number of requests (300 premium/month) |
| Windsurf | daily quota | $20/month | $200(Max) | multi-model | Daily/weekly quota system |
| Claude Code | - | $20/month (Pro) | $100(5x)/$200(20x) | Opus 4.6, Sonnet 4.6 | Token Budget (5 hours rolling) |
| Cline | completely free | - | - | BYOM (comes with key) | Direct by API price |
| Augment Code | - | $20/month | $60/month | own model | 40 K credits/month |
Bypass Subscriptions: The Real Cost of Per Volume APIs
If your needs go beyond "sitting in an IDE and writing code"-for example, if you want to build an automated Code Review Bot, do code analysis in CI/CD, or any non-human interaction scenarios-the Coding Plan won't work.
At this time, directly adjust the API and pay according to the volume. API prices in 2026 are already ridiculously cheap:
| model | Enter/M tokens | Output/M tokens | Suitable for the scene |
|---|---|---|---|
| DeepSeek-V3.2 (chat) | $0.28 | $0.42 | Common dialogue/code |
| DeepSeek-V3.2 (reasoner) | $0.28 | $0.42 | Complex reasoning/architecture |
| Claude Sonnet 4.6 | $3.00 | $15.00 | Highest quality code |
| Claude Sonnet 4.6 (Batch) | $1.50 | $7.50 | batch processing |
| Claude Haiku 4.5 | $1.00 | $5.00 | Lightweight/Classification |
| Gemini 2.5 Pro | free | free | 2-5 RPM,~100 RPD |
| Gemini 2.5 Flash | free | free | 15 RPM,~1500 RPD |
| OpenRouter DeepSeek R1 | free | free | 20 RPM, 200 RPD |
| Qwen 3-Max (Alibaba Cloud) | ¥2/M | ¥6/M | Chinese optimization |
Example scenario: A team runs daily automated code review ****
Assumption: 50 PRs per day, and an average of 2000 lines of code per PR need to be reviewed
- Each PR approximately 15 K input tokens +3 K output tokens
- 50 PR/day × 30 days = 750 K input +150 K output/month
| programme | monthly cost |
|---|---|
| DeepSeek V3.2 | ¥1.5 |
| Claude Sonnet Batch | ¥15 |
| Gemini 2.5 Flash | ¥0 |
| GPT-4o | ¥45 |
Coding Plan is completely inappropriate in this scenario (automated scripts are prohibited), but the monthly cost of the volume-based API is less than the cost of a cup of milk tea.
Who should buy a Coding Plan? Who shouldn't?
People suitable for the Coding Plan: ****
- writes code in the IDE for 4 + hours a day
- wants to switch between Qwen/GLM/DeepSeek at any time to see the effect
- cannot stand the uncertainty of pay-per-volume
- does not require integrating AI capabilities into other systems
People not suitable for the Coding Plan: ****
- wants to build automated workflows (CI/CD code review, automated test generation)
- wants to embed AI into its product backend
- requires a huge amount of Opus models (it is more cost-effective to go directly to Claude Code Max)
- is a part-time developer who occasionally writes code (free layer is enough)
ultimate price/performance ranking
Based on model quality, quota, tool compatibility, and stability, I give my ranking:
Domestic Coding Plan (Programming Assistance Scenario)
- Alibaba Cloud Refining Pro -The most complete models (8 models), the widest tool compatibility (20 +), ¥ 39.9 for the first month. Comprehensive first, no suspense
- Volcano Ark Pro -Bean Bag 2.0 Code model has unique advantages in Agent scenarios and supports DeepSeek V3.2
- MiniMax ¥ 119/month -No weekly quota is a killer, suitable for sprint concentrated explosive development
- Tencent Cloud Pro -It is almost the same as Ali, and has a slightly better understanding of the Mixed Origin Model in Chinese. Choose Tencent or Ali to see your cloud ecosystem binding
- Unquestioned Core Dome ¥ 19.9/month -Ultimate low-cost long-term holding solution
Overseas Tools (Product Experience Scenarios)
- Cursor Pro $20/month -The strongest IDE experience currently, Composer multi-file editing is invincible
- Claude Code Pro $20/month -Terminal Party First Choice, Million Tokens Context, Deep Understanding of the entire codebase
- GitHub Copilot Pro +$39/month -Top choice for corporate compliance, IP compensation + audit logs
API Volume Solution (Automation/Back-end Scenario)
- DeepSeek V3.2 official -$0.28/$0.42, performance 90% GPT-5, cost 1/50
- Claude Batch API -Highest quality-50% price, suitable for offline batch tasks
- Gemini 2.5 Flash free -zero cost, 15 RPM is lightweight enough for scenarios
- OpenRouter free model -Qwen 3 Coder 480B, DeepSeek R1 free call
's most rational entry strategy
If you are a full-time developer : Use Alibaba Cloud Refining Lite first month ¥ 7.9 and Tencent Cloud Lite first month ¥ 7.9, each of which will experience it for one month. Choose a renewal Pro in the second month. Cursor Pro or Claude Code Pro are also reserved for overseas tools.
If you want to do automated integration : Forget the Coding Plan, register directly for the DeepSeek platform (free 5 million Tokens), or use the Claude Batch API. The cost is extremely low and will not be banned.
If you want to spend the least amount of money : Cline (open source free)+ OpenRouter free model + Gemini 2.5 Flash free API. Monthly cost ¥ 0, quality 80 points.
If your budget is not a problem : Claude Code Max 20x ($200/month), one tool solves all problems. It understands your entire warehouse in the terminal, 1M token context, Opus 4.6 quality.
written at the end
The AI programming tool market in 2026 has passed the "novelty" stage and entered the real "who can help me work?" stage. The essence of the Coding Plan is to transform "fear of quantity" into "fixed peace of mind"-but it also locks you in the IDE interaction scenario.
My judgment is: Coding Plan is suitable for the scenario of "people sitting in front of the computer writing code", and API is suitable for the scenario of "letting the machine run for you". The two are not substitutes, but complementary. The most efficient developers often hold a Coding Plan (daily programming)+ an API account (automation pipeline) at the same time.
As for which one to choose-the price war has reached this level, the gap is no longer in terms of money. The gap lies in model quality (GLM-5 is currently the strongest domestically produced), tool compatibility (Ali is the most extensive), and stability (Volcanic Ark was exploded and oversold).
Spending ¥ 7.9 for a month is more useful than reading a hundred review articles.